Intensity of Relationship Between Words: Using Word Triangles in Topic Discovery for Short Texts

Uncovering latent topics from given texts is an important task to help people understand excess heavy information. This has caused the hot study on topic model. However, the main texts available daily are short, thus traditional topic models may not perform well because of data sparsity. Popular models for short texts concentrate on word co-occurrence patterns in the corpus. However, they do not consider the intensity of relationship between words. So we propose the new way, called word-network triangle topic model (WTTM). In WTTM, we search for the word triangles to measure the relations between words. The results of experiments on real-world corpus show that our method performs better in several evaluation ways.